CN108037995A - Distributed electromagnetic situation simulation computing system based on GPU - Google Patents
Distributed electromagnetic situation simulation computing system based on GPU Download PDFInfo
- Publication number
- CN108037995A CN108037995A CN201711174176.9A CN201711174176A CN108037995A CN 108037995 A CN108037995 A CN 108037995A CN 201711174176 A CN201711174176 A CN 201711174176A CN 108037995 A CN108037995 A CN 108037995A
- Authority
- CN
- China
- Prior art keywords
- data
- gpu
- computing
- task
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
Abstract
A kind of distributed electromagnetic situation simulation computing system based on GPU disclosed by the invention, there is provided a kind of computation capability that can obtain higher, the flexibility of higher and robustness, more inexpensive Electromagnetic Situation simulation computing system.The technical scheme is that:Integrated data management layer reads the basic data of Electromagnetic Situation simulation calculation needs from file system:Device data, environmental data and terrain data, formation operational data is pushed to distributed real-time operation layer and is handled after basic data is integrated.The computing resource that distributed real-time operation layer calls the GPU in hardware device level to calculate card and CPU board card handles operational data, handling result data return to integrated data management layer and form electromagnetic data, battle state display layer is given by battle state display data-pushing is formed after basic data and electromagnetic data synthesis, battle state display data are sent into GPU display cards, the visualization of situation data is carried out, and human-computer interaction interface is provided.
Description
Technical field
The present invention proposes a kind of distributed electromagnetic situation simulation computing system based on GPU.
Background technology
Radio electromagnetsm situation simulation and analysis has the function that to hold the balance in wireless communication field, it is each by studying
The signal model of class electromagnetic equipment, electromagnetic wave propagation model and consider influence of all kinds of interference to Electromagnetic Wave Propagation, make
It can accurately predict electromagnetic wave propagation distance, be disturbed situation and the electric wave coverage condition in certain region etc..Only can
Image, describe Electromagnetic Situation and efficient and rational radio-frequency spectrum could be managed and be utilized exactly.Due to requiring to simulate
Communication and other electromagnetic equipment huge numbers, thereby increases and it is possible to enormous amount, radiation efficiency by weather, landform, equipment or target transport
Dynamic rail mark etc. influences, while to consider the interference between different radiation sources so that the computation complexity of Electromagnetic Situation emulation is very
Height, common computer cannot meet the needs of Electromagnetic Situation emulation at all.With the howling success of business cloud computing, using also getting over
Come more extensive.Cloud computing technology is connected substantial amounts of computing device by network, virtually turns to resource pool, and can be carried out certainly
My maintenance and management.Calculating task can be intelligently distributed on the resource pool that a large amount of computers are formed by cloud computing, make user
Powerful computing power, memory space and information service can be obtained.Therefore, can be that Electromagnetic Situation emulates using cloud computing technology
Calculate and the support of the abilities such as powerful calculating, storage is provided.
Traditional Electromagnetic Situation simulation calculation is realized using CPU mostly.Due to the restriction of CPU architectures, it is calculated
Ability can not meet the growth expectations of " Moore's Law " already, and the check figure that multi-core CPU can accommodate is also extremely limited, this causes
Parallel artificial based on CPU is difficult to the growth requirement for meeting future radios Electromagnetic Situation system emulation analysis.At the same time, by
In the restriction of hardware condition, traditional Electromagnetic Situation parallel algorithm is difficult the breakthrough for having substance in computational efficiency.In recent years,
Important hardware foundations of the GPU (Graphics Processing Unit) as traditional graph process field parallel computation, gradually
Powerful calculating potentiality are shown in general scientific algorithm field, its calculated performance exceedes CPU of the same period already.Meanwhile by
It is integrated in large number of calculating core on one piece of GPU card, the parallel computation based on GPU is not only cheap, floor space
It is small and more energy efficient.Performance advantage of the design concept of GPU for Large-scale parallel computing in the application of big handling capacity, i.e. bigger
The Memory (i.e. video memory) of bandwidth, will obtain more excellent than the simple performance that parallel computation is carried out using CPU.
Existing Electromagnetic Situation analogue system framework is calculated based on CPU more, since the calculating core of CPU is less, controls mould
Block and caching occupy most of space of chip, be adapted to processing degree of parallelism is low, data locality is obvious, it is complicated, include
The computational problem of a large amount of branched structures.GPU is then formed by largely calculating core, and the ratio very little of control module and caching, is fitted
Conjunction processing operation strength is big, degree of parallelism is high, control flow is simple.And the working frequency of GPU cores is than traditional CPU faster,
So allow for GPU many more powerful than the disposal ability of single cpu.In recent years, GPU is because its huge calculating potentiality is in general meter
Calculation field just receives more and more attention, and has scholar's prediction, and the parallel computation based on GPU represents following high-performance calculation
Development trend.From comparing the characteristics of GPU and CPU:It is most of in CPU that (general 70%) transistor is used for building Cache also
Some control unit, part that responsible logic counts is simultaneously few, and it is exactly a huge computing array that GPU is whole, because
CPU will be far longer than by being responsible for the part counted of logic in this GPU, in logical operation it is stronger than CPU very much;And GPU is to Cache
Dependence it is smaller than CPU because GPU's is high speed bus, and CPU needs to handle different types of data, with greater need for storage
The cooperation of device instructs to perform load/store, therefore by being used as on-chip memory in one Cache of CPU internal builds, matches somebody with somebody
DRAM, ROM whole memory architecture are closed to coordinate CPU to handle load/store instructions;More also it is total linear speed of CPU
Degree falls behind very much.In addition, GPU needs that does to be operated in complexity and can not show a candle to CPU, do not have between GPU data to be treated
Any association, it is possible to parallel to perform.To sum up, CPU design original intention is more suitable for serial computing, and GPU is adapted on a large scale simultaneously
Row calculate, therefore using GPU computation capability can greatly improve Electromagnetic Situation emulate calculating speed.
More companies have developed many outstanding electromagnetic field simulation softwares both at home and abroad at present, for example, MWS, Mafia and
HFSS etc., these softwares have the advantages that easy to operate, powerful, become strong instrument of being engaged in scientific research.But these
Software can only be run on unit mostly, seem unable to do what one wishes when individually handling large-scale electromagnetism and calculating.Even some softwares
Support distributed arithmetic, but also can only be that single calculating task is handled, the batch that can not complete multitask calculate and
And price is very expensive, cost performance is not high.In order to calculate the time using distributed parallel mechanism to reduce, people are frequently with MPI
Calculated etc. mechanism establishment specific procedure.But at this time, although calculating the time reduces, programming time adds, always
Development time may not reduce, and algorithm comparison is single, and versatility is also poor.And this method can not also be supported to use
Family, multi-job operation, limit the performance of distributed type assemblies, can not realize the efficiently shared of resource.In addition, distributed system passes through
Frequently with some job scheduling systems, such as PBS softwares.But these job scheduling systems directly melt with general electromagnetic computing software
Close relatively difficult.As it can be seen that in Electromagnetic Calculation field, calculating task is typically complex, it is necessary to expend substantial amounts of computer money
Source, only can be taken too long with single computer, seriously affect computational efficiency, therefore need the support of distributed type assemblies.Cloud computing skill
Art all helps fault-tolerant ability of the intelligent management of resource, the intelligent scheduling ability of task and cloud computing distributed system etc.
Efficient in networking computing resource utilizes, and can be emulated for Electromagnetic Situation and provide more powerful computing capability, meet its demand.And
The characteristics of cloud computing technology provides cloud service with unified interface, interface, also beneficial to the development and maintenance of Electromagnetic Simulation system.
The requirement of real-time of Electromagnetic Situation analogue system.The cloud computing technology such as most successful Hadoop of business application is not at present
Meet the requirement of real-time of application.That although full dose data processing uses is famous hadoop or hive mostly, as one
A batch processing system, hadoop have obtained widely making with the advantages that its handling capacity is big, automatic fault tolerant in mass data processing
With.But hadoop is bad to calculate in real time, because its naturally life for batch processing, this, which is also that industry is consistent, is total to
Know.Otherwise s4, storm and puma these real time computation systems are there will not be within nearest 2 years to emerge.And Hadoop etc. is current
Although most successful cloud computing technology handling capacity is high, possesses the disposal ability of mass data, it the shortcomings that and advantage similarly
It is distinct:Delay is big, response is slow, O&M is complicated, is not particularly suited for the high calculating application environment of requirement of real-time.
The content of the invention
In order to make up the shortcoming of current Electromagnetic Situation analogue system, can be reduced the object of the present invention is to provide one kind whole
A system degree of coupling, obtains the computation capability of higher, the flexibility of higher and robustness, more inexpensive to meet Electromagnetic Situation
The distributed electromagnetic situation simulation computing system based on GPU computing clusters of analogue system requirement of real-time.
The above-mentioned purpose of the present invention is reached by following measures.A kind of distributed electromagnetic situation emulation meter based on GPU
Calculation system, including:Based on model-view-controller (MVC, Model-View-Controller) frame, from top to bottom successively
It is divided into:Battle state display layer, integrated data management layer, distributed real-time operation layer and hardware device level, four-layer structure, its feature exist
In:Integrated data management layer reads the basic data of Electromagnetic Situation simulation calculation needs from file system:Device data, environment
Data and terrain data, formation operational data is pushed to distributed real-time operation layer and is handled after basic data is integrated.
Distributed real-time operation layer calls GPU to calculate card and the computing resource of CPU board card handles operational data, the knot after processing
Fruit data return to integrated data management layer and form electromagnetic data;Integrated data management layer integrates basic data and electromagnetic data
Battle state display data-pushing is formed afterwards gives battle state display layer;Battle state display data are sent into GPU display cards by battle state display layer, are carried out
The visualization of situation data, and human-computer interaction interface is provided.
The beneficial effect of present invention method compared with prior art is:
The degree of coupling of whole system can be reduced.The present invention is based on model-view-controller (MVC, Model-View-
Controller) frame, is designed with reference to GPU and cloud computing technology, is separated with a kind of service logic, data and interface display
Method tissue code, service logic is gathered inside system unit, there is provided possess the electromagnetism state of real-time computing
Gesture simulation computing system framework, while improving and personalized customization interface and user mutual, it is not necessary to rewrite business
Logic.The advantage being designed using MVC frameworks is to be separated data transfer with data processing, data management, data
Processing is shown with data to be separated, and can be reduced the degree of coupling of whole system, be improved the robustness of system.
The computation capability of higher can be obtained.The present invention is according to the characteristics of Electromagnetic Situation simulation computing system, based on storm points
The real-time Computational frame of cloth is designed the computation model, Portable Batch System and calculation process of system.Based on model-regard
System, is divided into battle state display layer, comprehensive number by figure-controller (MVC, Model-View-Controller) frame from top to bottom
According to the four-layer structure of management level, distributed real-time operation layer and hardware device level, using GPU as core is calculated, with the side of data flow
Formula carries out Electromagnetic Situation data calculating.System resource is carried out mutually based on the framework of cloud computing on the basis of this layered approach
Connection and management, have put into practice the separation that data management, data processing and data are shown, have been greatly improved the parallel computation energy of system
Power, meets computation complexity and the real-time computing requirement of Electromagnetic Situation analogue system.
System administration, the intelligence of task scheduling and the flexibility of higher and robustness.The present invention uses MVC architecture designs
Software reuse it is high, life cycle cost is low, deployment is fast, maintainability is high.Integrated data management layer reading disk or network
The data of access, dispatch computing system resource, are transmitted device data, environmental data and terrain data by external interface
Data operation and processing are completed to distributed real-time operation layer.Using cloud computing technology to the intelligent management of resource, to task
Intelligent scheduling ability, the resource management of system and Utilization ability will surmount common computer networking significantly.Without considering network
In the case of being transmitted as bottleneck, no matter the size and complexity of task, can be reached by way of increasing calculate node quantity
To the real-time of tasks carrying.Based in the distributed clouds of Computational frame in real time of storm, each calculate node is ill-mannered
State, therefore each calculate node can be substituted by any other node, possess the node homogeneity in Distributed Architecture.Calculate
During, according to the demand of computing resource, the calculate node in frame can be increased or decreased arbitrarily, also can be in tasks carrying mistake
Added at any time or in journey since the other reasons such as failure exit, these operations do not interfere with the implementing result of task, carry significantly
The flexibility of high system and robustness.In addition, cloud computing service resource is accessed with the characteristic that interface is unified, interface is unified,
Also it is beneficial to the exploitation and maintenance of Electromagnetic Simulation system algorithm.
Lower cost meets the requirement of real-time of Electromagnetic Situation simulation calculation.The distributed real-time operation layer profit of the present invention
System hardware device resource is scheduled and managed with the resource management techniques of cloud computing, while the exception for providing system is caught
Obtain, log management, running state monitoring, subsystem management and control, task scheduling, the decision-making of algorithm burst, algorithmic procedure management and data
The service such as integration;Integrated data management layer, will from basic datas such as file system reading database, local file and telefiles
The electromagnetic data processing result of distributed real-time operation layer feedback is sent to battle state display layer after being integrated, and completes Electromagnetic Situation
Information visuallization.The different resource of numerous abilities and equipment are subjected to intelligent interconnection using cloud computing technology, efficiently cooperateed with
Complete to calculate, the high-performance server expensive without individually purchase is as computing resource.Based on the distributed real-time calculation blocks of storm
In the cloud of frame, each calculate node can be substituted by any other node, possess the node homogeneity in Distributed Architecture,
Therefore the computing resource with different characteristics and performance can become the calculate node of cloud.Meanwhile it is being without considering network transmission
In the case of bottleneck, no matter the size and complexity of task, can reach task by way of increasing calculate node quantity
The real-time of execution.
Brief description of the drawings
Fig. 1 is the distributed electromagnetic situation simulation computing system layering schematic diagram based on GPU.
Fig. 2 is the data flowchart of Fig. 1 embodiments.
Fig. 3 is the hardware structure diagram of Fig. 1 distribution real-time operation layers.
Fig. 4 is the work flow diagram of Fig. 3 distribution real-time operation layers.
This method is further illustrated with reference to the accompanying drawings and detailed description.
Embodiment
Refering to Fig. 1.In embodiment described below, a kind of distributed electromagnetic situation simulation computing system based on GPU,
Including:Based on model-view-controller (MVC, Model-View-Controller) frame, it is divided into from top to bottom:State
Gesture display layer, integrated data management layer, distributed real-time operation layer and hardware device level, four-layer structure, it is characterised in that:It is comprehensive
Data management layer reads the basic data of Electromagnetic Situation simulation calculation needs from file system:Device data, environmental data and
Terrain data, formation operational data is pushed to distributed real-time operation layer and is handled after basic data is integrated.It is distributed
Real-time operation layer calls the GPU in hardware device level to calculate card and the computing resource of CPU board card handles operational data, place
Result data after reason returns to integrated data management layer and forms electromagnetic data;Integrated data management layer is by basic data and electromagnetism
Battle state display data-pushing is formed after aggregation of data and gives battle state display layer;Battle state display layer shows battle state display data feeding GPU
Show card, carry out the visualization of situation data, and human-computer interaction interface is provided.
Battle state display layer is obtained from integrated data management layer needs landform, environment, deployed with devices and electromagnetic field letter to be shown
The battle state display data such as breath are visualized, there is provided human-computer interaction interface, display area adjustment, the mirror of Electromagnetic Situation analogue system
Head switching, focus device switching, and the function of the log information such as the alarm of system, debugging, operation.
Integrated data management layer reads Electromagnetic Situation analogue system from file system by external interface and calculates required basic number
According to including device data, environmental data and terrain data etc., distributed real-time operation layer is sent into after basic data is integrated and is obtained
The data processed result that must be fed back, and by external interface transmitted basic data and electromagnetic data complete to battle state display layer
Into Electromagnetic Situation information visuallization.
Distributed real-time operation layer is scheduled and managed to system hardware device resource using the resource management techniques of cloud computing,
Exception catching, log management, running state monitoring, subsystem management and control, task scheduling, the algorithm burst for providing system at the same time are determined
Plan, algorithmic procedure management and Data Integration etc. service.
Calculating processing, data exchange and the battle state display of the hardware supported whole system of hardware device level.Hardware device level includes:
CPU board card, multiple GPU calculate card and GPU display cards.CPU board card is the hard core control hardware cell of system, supports the pipe of system
Reason, scheduling of resource and a small amount of calculating task.GPU calculates the calculating core that card is system, supports the most data processing of system
Task.GPU boards are the display cores of system, support the display of final Electromagnetic Situation data.CPU board card passes through disk or network
Transmission mode obtains data, and dispatches GPU and calculate the most calculating task of card execution, finally by GPU display cards to synthesis
Calculation result data carry out terrain rendering, the visual work such as battle state display.Answered with reference to current business demand and follow-up system
Miscellaneous degree extension, the selection of hardware device are contemplated that reserved larger resource redundancy.
Refering to Fig. 2.The data flow of distributed electromagnetic situation simulation computing system based on GPU can be divided into three classes:Basic number
According to, operational data and battle state display data.Basic data refers to that Electromagnetic Situation calculates required basic information data, main bag
Include:Device data, environmental data and terrain data.Operational data refers to that integrated data management layer is sent into distributed real-time operation
Layer, the intermediate data flowed in operation layer (result data is a part for operational data);Battle state display data refer to comprehensive number
It is according to the integrated data management module of management level that basic data in distributed electromagnetic situation analogue system and electromagnetic data progress is whole
Close, be transferred to battle state display layer and be used for the visual data of situation information.Integrated data management layer is Electromagnetic Situation analogue system
Data management core, read the basic data that Electromagnetic Situation simulation calculation needs, including device data, environment from file system
Data and terrain data;Calculative Electromagnetic Situation data (i.e. operational data) are pushed to distributed real-time operation layer to carry out
After processing, operation result data return to integrated data management layer and form electromagnetic data.Integrated data management layer is by basic data
The battle state display data-pushing formed after being integrated with electromagnetic data gives battle state display layer.Battle state display layer send battle state display data
Enter GPU and show that card is visualized.Wherein, distributed real-time operation layer can be according to the characteristics of data and task, by most of fortune
Calculation task is pushed to GPU operation clusters, and small part processor active task is pushed to CPU.GPU operation clusters and CPU are by data operation knot
Fruit feeds back to distributed real-time operation layer.
Refering to Fig. 3.The main CPU board by being connected through High speed rear panel of the hardware configuration of this example distribution real-time operation layer
Card, multiple GPU calculate the composition such as board and SSD high speed solid hard-disc storage boards.CPU board card is the core tube of whole operation layer
Manage unit, support the upper and lower fulgurite of other boards manage, a key open and close machine with restart function.CPU board card can by read hard disk,
Network interface etc. carries out the data interaction of Local or Remote, by operating in operating system and management software reality on CPU board card
The scheduling of existing GPU computing clusters.GPU calculates the calculating core that board is whole operation layer, and GPU selects high performance core list
Member, the video memory of large capacity are simultaneously connected by high-speed bus, improve the data throughput between GPU and outside and GPU computing units
Bandwidth, helps to improve the parallel processing capability of whole GPU computing clusters.SSD high speed solid hard disks are the interfaces being locally stored,
Using the high-speed read-write performance of SSD solid state hard discs, the performance that local data reads and stores can be improved.The up to network of 10G connects
Mouth expansion board, the network interaction bandwidth of high speed is provided for system, the transmittability of network data can be increased substantially, help to carry
The transmittability of the high pending file of network and the extension of distributed arithmetic layer, meet that high speed distributed computing system is saved to calculating
The high request of network bandwidth between point.
Referring to Fig. 4.Distributed real-time operation layer mainly includes:Task scheduling modules, configuration management module, service distribution mould
Block, state management module, management and control service module, GPU computing clusters and CPU computing modules.It is comprehensive in distributed real-time operation layer
Close data management layer and initiate task requests to distributed real-time operation layer.Task scheduling modules receive appointing for integrated data management layer
Business request, and ask resource to configuration management module.Configuration management module is to appoint according to the type of task requests and operational data
Scheduler module of being engaged in distribution computing resource and algorithm.Task scheduling modules, can burst task after computing resource and algorithm is obtained
Be sent to service distribution module, by can not burst task be transmitted directly to the computing unit or CPU computing modules of GPU computing clusters
Computing unit.Service distribution module receive an assignment scheduler module burst task list obtain processor active task, according to task
Computing requirement, the computing unit distribution processor active task bag of computing unit or CPU computing modules to GPU computing clusters.At the same time will
Task run instruction is sent to GPU computing clusters or CPU computing modules, gathers the day of GPU computing clusters or CPU computing modules
Will, and instruction and daily record are synchronously reported to state management module;GPU computing clusters and CPU computing modules receive processor active task
Afterwards, asked according to the computing of service distribution module, the corresponding task bag of dynamic load simultaneously performs.Computing after the completion of tasks carrying
Data pass through Data Integration, return to integrated data management layer;State management module collects the speed of service of service distribution module
Information, is managed the state of service distribution module, and the session that will monitor GPU computing clusters and CPU computing modules in real time
Status information is sent into management and control service module.Management and control service module calculates service distribution module, GPU by human-computer interaction interface
The status information of cluster and CPU computing modules processing task is shown.The status information of processing task includes processing task bag
Total amount, current task speed packet and currently processed task type etc..
Distributed real-time operation layer is the calculating core of situation analogue system, according to task and data type, the meter of data
Calculation task will mainly be given the powerful GPU computing clusters of parallel processing capability and be carried out, and small part task is given at CPU computing modules
Reason.In the GPU computing clusters and CPU computing modules of the present embodiment, each computing unit, that is, calculate node be it is stateless,
Therefore any calculate node can be substituted by other nodes, meet the node homogeneity in Distributed Architecture.In addition, the present embodiment
In calculate node can arbitrarily increase or decrease, also can add or exit at any time in task implementation procedure, these operations are equal
The implementing result of task is not interfered with.In the case where being bottleneck without considering network transmission, the execution efficiency of task is saved with calculating
Point quantity it is directly proportional, therefore no matter the size and complexity of task, can by increase the quantity of calculate node reach appoint
The real-time that business performs.
Any feature disclosed in description of the invention, including any attached claim, summary and attached drawing, except non-specifically chatting
State, can alternative features equivalent by other or with similar purpose replaced.
Claims (10)
1. a kind of distributed electromagnetic situation simulation computing system based on GPU, including:Based on model-view-controller (MVC,
Model-View-Controller) frame, is divided into from top to bottom:Battle state display layer, integrated data management layer, distribution
Real-time operation layer and hardware device level, four-layer structure, it is characterised in that:Integrated data management layer reads electromagnetism from file system
The basic data that situation simulation calculation needs:Device data, environmental data and terrain data, form after basic data is integrated
Operational data is pushed to distributed real-time operation layer and is handled;Distributed real-time operation layer calls the GPU in hardware device level
The computing resource for calculating card and CPU board card handles operational data, and the result data after processing returns to integrated data pipe
Manage layer and form electromagnetic data;Integrated data management layer will form battle state display data-pushing after basic data and electromagnetic data synthesis
Give battle state display layer;Battle state display data are sent into GPU display cards by battle state display layer, carry out the visualization of situation data, and carry
For human-computer interaction interface.
2. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:Situation is shown
Show that layer obtains the battle state display number of need landform to be shown, environment, deployed with devices and electromagnetic field information from integrated data management layer
According to being visualized, there is provided the human-computer interaction interface of Electromagnetic Situation analogue system, display area adjustment, Shot change, focus are set
Standby switching, and the alarm of system, debugging, the log information of operation.
3. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:Comprehensive number
The basic data read according to management level by external interface from file system needed for the calculating of Electromagnetic Situation analogue system includes equipment
Data, environmental data and terrain data, are sent into distributed real-time operation layer and obtain the data of feedback after basic data is integrated
Handling result, and basic data and electromagnetic data are transmitted to battle state display layer by external interface and complete Electromagnetic Situation letter
The visualization of breath.
4. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:It is distributed
Real-time operation layer is scheduled and managed to system hardware device resource using the resource management techniques of cloud computing, while provides system
Exception catching, log management, running state monitoring, subsystem management and control, task scheduling, the decision-making of algorithm burst, the algorithmic procedure of system
Management and data integrated service.
5. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:Hardware is set
Standby layer includes:CPU board card, multiple GPU calculate card and GPU display cards;CPU board card is the hard core control hardware cell of system, branch
Hold management, scheduling of resource and a small amount of calculating task of system;GPU calculates the calculating core that card is system, supports that system is most of
Data processing task;GPU display cards are the display cores of system, support the display of final Electromagnetic Situation data;CPU board cartoon
Cross disk or network transmission mode obtains data, scheduling GPU calculates card and performs most calculating task, shown finally by GPU
Show that card carries out the visual works such as terrain rendering, battle state display to comprehensive calculation result data.
6. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:CPU board card
By read hard disk, network interface carry out Local or Remote data interaction, by operate in the operating system on CPU board card and
Management software realizes the scheduling of GPU computing clusters.
7. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:It is distributed
Real-time operation layer includes:Task scheduling modules, configuration management module, service distribution module, state management module, management and control service mould
Block, GPU computing clusters and CPU computing modules;Task scheduling modules receive the task requests of integrated data management layer, and to configuration
Management module asks resource;Configuration management module is task scheduling modules distribution meter according to the type of task requests and operational data
Calculate resource and algorithm;Task scheduling modules after computing resource and algorithm is obtained, can burst task be sent to service distribution mould
Block, by can not burst task be transmitted directly to the computing unit of GPU computing clusters or the computing unit of CPU computing modules.
8. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:Business is sent out
Cloth module receive an assignment scheduler module burst task list obtain processor active task, according to task computing requirement, to GPU calculate
The computing unit of cluster or the computing unit distribution processor active task bag of CPU computing modules, while task run instruction is sent to
GPU computing clusters or CPU computing modules, gather the daily record of GPU computing clusters or CPU computing modules, and will instruct same with daily record
Step is reported to state management module.
9. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:GPU is calculated
After cluster and CPU computing modules receive processor active task, asked according to the computing of service distribution module, the corresponding task of dynamic load
Wrap and perform task, the operational data after the completion of tasks carrying passes through Data Integration, returns to integrated data management layer.
10. the distributed electromagnetic situation simulation computing system based on GPU as claimed in claim 1, it is characterised in that:State pipe
The speed of service information of module collection service distribution module is managed, the state of service distribution module is managed, and will supervise in real time
The session state information for controlling GPU computing clusters and CPU computing modules is sent into management and control service module;Management and control service module passes through man-machine
Interactive interface, is shown the status information of service distribution module, GPU computing clusters and CPU computing modules processing task.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711174176.9A CN108037995A (en) | 2017-11-22 | 2017-11-22 | Distributed electromagnetic situation simulation computing system based on GPU |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711174176.9A CN108037995A (en) | 2017-11-22 | 2017-11-22 | Distributed electromagnetic situation simulation computing system based on GPU |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108037995A true CN108037995A (en) | 2018-05-15 |
Family
ID=62094187
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711174176.9A Pending CN108037995A (en) | 2017-11-22 | 2017-11-22 | Distributed electromagnetic situation simulation computing system based on GPU |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108037995A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115904738A (en) * | 2023-01-05 | 2023-04-04 | 摩尔线程智能科技(北京)有限责任公司 | Management system and control method for data processing device cluster |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102799708A (en) * | 2012-06-18 | 2012-11-28 | 苏州超集信息科技有限公司 | Graphic processing unit (GPU) high-performance calculation platform device applied to electromagnetic simulation |
CN102981064A (en) * | 2012-10-09 | 2013-03-20 | 中国人民解放军63892部队 | Aircraft external radio frequency electromagnetic environment prediction method and prediction system |
CN105049485A (en) * | 2015-06-09 | 2015-11-11 | 中国石油大学(华东) | Real-time video processing oriented load-aware cloud calculation system |
EP3023873A1 (en) * | 2014-11-24 | 2016-05-25 | Samsung Electronics Co., Ltd. | Electronic device and method for providing map service |
US20170288980A1 (en) * | 2016-04-05 | 2017-10-05 | Bit Lion, LLC | System for visualization of electromagnetic wave distribution and collaborative web-based design for optimal distribution of emitters |
-
2017
- 2017-11-22 CN CN201711174176.9A patent/CN108037995A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102799708A (en) * | 2012-06-18 | 2012-11-28 | 苏州超集信息科技有限公司 | Graphic processing unit (GPU) high-performance calculation platform device applied to electromagnetic simulation |
CN102981064A (en) * | 2012-10-09 | 2013-03-20 | 中国人民解放军63892部队 | Aircraft external radio frequency electromagnetic environment prediction method and prediction system |
EP3023873A1 (en) * | 2014-11-24 | 2016-05-25 | Samsung Electronics Co., Ltd. | Electronic device and method for providing map service |
CN105049485A (en) * | 2015-06-09 | 2015-11-11 | 中国石油大学(华东) | Real-time video processing oriented load-aware cloud calculation system |
US20170288980A1 (en) * | 2016-04-05 | 2017-10-05 | Bit Lion, LLC | System for visualization of electromagnetic wave distribution and collaborative web-based design for optimal distribution of emitters |
Non-Patent Citations (2)
Title |
---|
延伟勤等: "基于HLA的分布式雷达系统仿真研究", 《大众科技》 * |
穆兰等: "空间电磁环境可视化系统的研究与应用", 《系统仿真学报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115904738A (en) * | 2023-01-05 | 2023-04-04 | 摩尔线程智能科技(北京)有限责任公司 | Management system and control method for data processing device cluster |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Rausch et al. | Optimized container scheduling for data-intensive serverless edge computing | |
Saxena et al. | A proactive autoscaling and energy-efficient VM allocation framework using online multi-resource neural network for cloud data center | |
JP7322273B2 (en) | Spiking Neural Network Computing Systems and Methods Used in Brain-Inspired Intelligence and Cognitive Computing | |
Kaur et al. | Container-as-a-service at the edge: Trade-off between energy efficiency and service availability at fog nano data centers | |
Fujimoto | Research challenges in parallel and distributed simulation | |
CN104898573B (en) | A kind of digital control system data acquisition based on cloud computing and processing method | |
US10802864B2 (en) | Modular reinforcement-learning-based application manager | |
DE102021209145A1 (en) | METHOD AND DEVICE FOR COORDINATING EDGE PLATFORMS | |
CN110389838A (en) | A kind of Real-Time Scheduling suitable for virtual resource and online migration management-control method | |
CN104486255B (en) | Service resources dispatching method and device | |
US9239734B2 (en) | Scheduling method and system, computing grid, and corresponding computer-program product | |
CN105978704A (en) | Creating new cloud resource instruction set architecture | |
CN110753107B (en) | Resource scheduling system, method and storage medium under space-based cloud computing architecture | |
CN109684074A (en) | Physical machine resource allocation methods and terminal device | |
US20160380862A1 (en) | Methods and systems to evaluate data center resource allocation costs | |
CN105373432B (en) | A kind of cloud computing resource scheduling method based on virtual resource status predication | |
CN104123182A (en) | Map Reduce task data-center-across scheduling system and method based on master-slave framework | |
CN103581339A (en) | Storage resource allocation monitoring and processing method based on cloud computing | |
US20230028934A1 (en) | Methods and decentralized systems that employ distributed machine learning to automatically instantiate and manage distributed applications | |
Kijsipongse et al. | A hybrid GPU cluster and volunteer computing platform for scalable deep learning | |
CN103595815A (en) | Method for distributing storage resources based on cloud computing | |
CN111814911A (en) | Electric power AI training platform based on containerization management and training method thereof | |
CN112631986A (en) | Large-scale DSP parallel computing device | |
CN108563787A (en) | A kind of data interaction management system and method for data center's total management system | |
Ataie et al. | Modeling and evaluation of dispatching policies in IaaS cloud data centers using SANs |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180515 |
|
WD01 | Invention patent application deemed withdrawn after publication |